Research Of Image Segmentation Based On Pulse Coupled Neural Network

In the military field, digital image feature extraction, object recognition, terrain contour matching navigation are more and more applied, and these applications are all based on image segmentation. In this paper,we focus on image segmentation based on pulse coupled neural network (PCNN).We describe the basic principles, operating mechanism and characteristics of the PCNN model and applications of the model in image processing in this paper. In order to solve the inherent problem that the model parameters are difficult to be determined, we bring up some solutions for those problem.The algorithm can set the model parameters adaptively according to the neighborhood pixels and average grayscale coefficient of the center pixel. According to the overall gray-scale of the images the algorithm of this paper can adaptively adjustment the threshold decay time constant, so the algorithm has strong adaptability. The algorithm is applied to segmented infrared images and aerial images which are low contrast images, and the simulation results show clear advantages. This paper introduces the idea of inhibitory input to optimize the composition of the internal activities, making the algorithm with a strong noise immunity and high accuracy of segmentation. The proposed algorithm also improved the connection input, using gray value directly as linking input instead of0,1pulse input, increasing the accuracy of the algorithm. In order to verify the superiority of the algorithm, we compare algorithm with the simplified PCNN and the maximum between-class variance (OTSU) respectively.In the end, we do some research on color image segmentation, indicating the necessity of color segmentation. Then we introduce several color space commonly used in image processing, and take improved PCNN algorithm into color image segmentation based on these color space respectively. In order to verify which color space is more fitting for image segmentation, we do a simulation comparison and analysis. The experimental results show that the effect of the improved algorithm in color image segmentation based on RGB space is better than HSI color space.